Object Detection

We recommend using these caffe models with py-RFCN-priv we are releasing the training code and files, the models and more experiments will come soon.

Object Detection Performance on PASCAL VOC. (More experiments)

1. Original Faster-RCNN training on VOC 2007+2012 trainval and testing on VOC 2007 test.

Network	mAP@50(%)	training speed	training memory	testing speed	testing memory
resnet18	70.02	9.5 img/s	1,235MB	17.5 img/s	989MB
resnet101-v2	79.6	3.1 img/s	6,495MB	7.1 img/s	4,573MB
wrn50-2	78.59	2.1 img/s	4,895MB	4.9 img/s	3,499MB
resnext50-32x4d	77.99	3.6 img/s	5,315MB	7.4 img/s	4,305MB
resnext101-32x4d	79.98	2.7 img/s	7,836MB	6.3 img/s	5,705MB
inception-v4	81.49	2.6 img/s	6,759MB	5.4 img/s	4,683MB
inception-resnet-v2	80.0	2.0 img/s (batch=112)	11,497MB	3.2 img/s	8,409MB
air101	81.0	2.4 img/s	7,747MB	5.1 img/s	5,777MB

• To reduce memory usage, we merge all the models batchnorm layer parameters into scale layer, more details please refer to faster-rcnn-resnet or pva-faster-rcnn;
• We also split the deploy file to rpn deploy file and rcnn deploy file for adopting more testing tricks.
• Performanc, speed and memory are calculated on py-RFCN-priv with Nvidia Titan pascal, we do not guarantee that the results can be reproduced under any other conditions;
• All the models are trained on a single scale (600*1000) with image flipping and train-batch=128 for 80,000 iterations, tested on the same single scale with test-batch=300 and nms=0.3;

2. Faster-RCNN-2fc-OHEM training on VOC 2007+2012 trainval and testing on VOC 2007 test.

Network	mAP@50(%)	training speed	training memory	testing speed	testing memory
se-inception-v2	(77.57)	9.4 img/s	2,453MB	15.9 img/s	1,573MB
se-resnet50	(79.73)	6.2 img/s	4,129MB	12.8 img/s	2,175MB
resnet101-v2	80.6(80.49)	5.0 img/s	5,833MB	10.5 img/s	3,147MB
air101	(81.47)	3.4 img/s	6,653MB	8.7 img/s	4,503MB
inception-v4-3x3	81.12(81.30)	3.73 img/s	5,383MB	10.1 img/s	3,217MB

• 2fc means: conv256d --- fc1024d --- fc1024d;
• The mAP@50 score in parentheses is training with ohem and multigrid;

3. RFCN-OHEM training on VOC 2007+2012 trainval and testing on VOC 2007 test.

Network	mAP@50(%)	training speed	training memory	testing speed	testing memory
resnet18	71.82	14.3 img/s	1,215MB	23.4 img/s	899MB
se-inception-v2	(78.23)	10.2 img/s	2,303MB	14.0 img/s	1,567MB
se-resnet50	(79.19)	6.3 img/s	3.999MB	11.7 img/s	2,205MB
resnet101-v2	78.93(79.9)	4.9 img/s	5,719MB	10.4 img/s	3,097MB
resnext101-32x4d	79.98(80.35)	3.8 img/s	6,977MB	8.8 img/s	4,761M
air101	79.42(80.93)	3.4 img/s	6,525MB	8.5 img/s	4,477MB
inception-v4	80.2	4.1 img/s	4,371MB	10.3 img/s	2,343MB

• The mAP@50 score in parentheses is training with ohem and multigrid;

Object Detection Performance on MSCOCO. (More experiments)

1. Results training on MSCOCO2017-trainval and testing on test-dev2017.

Network	mAP	mAP@50	mAP@75	mAP@S	mAP@M	mAP@L
RFCN-se-inception-v2 with ms-train & ohem & multigrid	32.6	53.6	34.5	12.5	35.1	48.4
RFCN-se-resnet50 with ms-train & ohem & multigrid	32.9	54.4	34.8	13.0	35.3	48.1
RFCN-air101 with ms-train & ohem & multigrid	38.2	60.1	41.2	18.2	41.9	53.0
Faster-2fc-air101 with ms-train & ohem & multigrid	36.5	60.4	38.1	15.5	39.5	53.5

• All the models are test on a single scale (600*1000) without any bells and whistles;

2. Context Pyramid Attention Network (CPANet) results training on MSCOCO2017-trainval and testing on test-dev2017.

Network	mAP	mAP@50	mAP@75	mAP@S	mAP@M	mAP@L
CPANet-air101 with ms-train & ohem & multigrid & 800-scale-test	41.9	64.8	45.5	24.0	45.9	54.6
CPANet-air101 with ms-train & ohem & multigrid & 800-scale-test & snms	42.7	65.4	46.7	24.6	46.8	55.6
CPANet-air101 with ms-train & ohem & multigrid & 800-scale-test & snms & flipping	43.5	65.9	47.5	25.1	47.7	56.6